if status refers to deference graph centrality, I’d argue that that variable needs to be fairly heavily L2 regularized so that the social network doesn’t have fragility. if it’s not deference, it still seems to me that status refers to a graph attribute of something, probably in fact graph centrality of some variable, possibly simply attention frequency. but it might be that you need to include a type vector to properly represent type-conditional attention frequency, to model different kinds of interaction and expected frequency of interaction about them. in any case, whatever attribute it is you’re modeling with the reduced “status” variable, I’d argue it’s probably not good to have status inequality and that effective use of knowledge of the interaction-pattern known as “status” is to identify people who don’t have enough in a given interaction and ensure they get some, conditional on their interaction-safety allowing it, or something. it’s probably not something where enhancing inequality is a good idea.
if status refers to deference graph centrality, I’d argue that that variable needs to be fairly heavily L2 regularized so that the social network doesn’t have fragility. if it’s not deference, it still seems to me that status refers to a graph attribute of something, probably in fact graph centrality of some variable, possibly simply attention frequency. but it might be that you need to include a type vector to properly represent type-conditional attention frequency, to model different kinds of interaction and expected frequency of interaction about them. in any case, whatever attribute it is you’re modeling with the reduced “status” variable, I’d argue it’s probably not good to have status inequality and that effective use of knowledge of the interaction-pattern known as “status” is to identify people who don’t have enough in a given interaction and ensure they get some, conditional on their interaction-safety allowing it, or something. it’s probably not something where enhancing inequality is a good idea.